Bladed servers are computing systems that provision servers or other computer resources on individual cards, or blades. There are many types of blades—server blades, storage blades, network blades, processor blades, etc. The blades of a server are typically housed together in a single structure, creating high-density systems with a modular architecture that ensures flexibility and scalability. Thus, bladed servers reduce space requirements. Server blades, along with storage, networking and other types of blades, are typically installed in a rack-mountable enclosure, or chassis, that hosts multiple blades that share common resources such as cabling, power supplies, and cooling fans.
The telecommunications industry has been using bladed server technology for many years. The condensed server bladed architecture also benefits people and businesses that use the Internet to generate revenues and provide services to customers, that are moving some of their business processes to the Web, and that need the flexibility to deploy Internet-edge applications in their own data center. Because of recent developments in technology, bladed servers are now used for applications such as web hosting, web caching and content streaming. Web caching, for example, stores frequently requested Web content closer to the user so objects can be retrieved more quickly, thereby reducing the time and the bandwidth required to access the Internet. Companies and individuals are now streaming media, such as video, audio, and interactive, in order to more effectively communicate both internally and externally. This has led to a massive growth of rich media content delivery on the Internet. Bladed servers are being used to meet these new demands.
Bladed servers, however, create challenging engineering problems, due largely in part to heat produced by the blades and limited space in the chassis. Typically, bladed server systems are limited by an underlying power and thermal envelope. For example, a chassis that hosts a bladed system may only be designed to utilize a limited number of watts. That is, the chassis can only can consume so much power and is limited in the amount of airflow that is available to cool the blades in the chassis.
Engineering challenges occur in optimizing the tradeoff between performance and thermal and power requirements. In a bladed architecture multiple blades, each representing a separate system, are present in the same chassis. Associated with the chassis are a specific set of power and thermal requirements. Specifically, these requirements put a limit on the amount of power that can be consumed by the respective blades.
Known power limiting strategies include powering down a CPU functional unit, e.g., a floating-point unit or an on-die cache, or trading off speed for reduced power consumption in a hard drive. Power limitations also put a constraint on the frequency that the processors on the blade can run, and thus, limits the performance. In addition, the processors in a system are usually all configured to operate at the same frequency. This further limits the ability for the individual processors to operate at optimal performance and capacity.
Prior solutions run all the blades at a performance level less than their maximum in order to meet the overall chassis power and thermal cooling budget. A disadvantage associated with this solution is that the performance of each blade is degraded or diminished to fall within these budgets. For example, if the ability of the chassis to cool is limited to X and there are Y blades, each blade can only contribute approximately X/Y to the dissipated power in the chassis. Thus, each blade is limited to the performance associated with an X/Y power level.
Another solution has been to add a plurality of loud, space-consuming fans that require expensive control circuitry. These cooling systems increase the cost of the server blade system, leave less space for other features within the chassis for other features, and run a higher risk for failures and increased downtime. Other solutions have included limiting the number of I/O cards in the system, as well as restricting the number of other use features. A further solution has been to reduce the power budget available for other features in the system.
What is needed is a method for modulating the frequency and voltage of a blade in a bladed architecture system in accordance with the performance demands of the individual blade and the overall system power and thermal budget of the system.